Delete race_audio/.ipynb_checkpoints

race_audio/.ipynb_checkpoints/filter-checkpoint.py

@@ -1,197 +0,0 @@
-#!/usr/bin/env python3
-"""
-根据音频时长过滤race_benchmark.json，生成新的benchmark文件
-仅保留音频时长在指定时间以上的QA对
-"""
-
-import json
-import os
-import librosa
-import soundfile as sf
-from collections import defaultdict
-import argparse
-
-def get_audio_duration(audio_path):
-    """获取音频文件的时长（秒）"""
-    try:
-        # 使用librosa获取音频时长（更快，只读取header）
-        duration = librosa.get_duration(path=audio_path)
-        return duration
-    except Exception as e:
-        try:
-            # 备选方案：使用soundfile
-            info = sf.info(audio_path)
-            duration = info.frames / info.samplerate
-            return duration
-        except Exception as e2:
-            print(f"无法读取音频文件 {audio_path}: {e2}")
-            return 0
-
-def format_duration(seconds):
-    """将秒数格式化为分:秒的形式"""
-    minutes = int(seconds // 60)
-    secs = int(seconds % 60)
-    return f"{minutes}:{secs:02d}"
-
-def filter_race_benchmark_by_duration(race_audio_dir, race_benchmark_path, min_duration_seconds=120):
-    """
-    根据音频时长过滤race_benchmark.json，生成新的benchmark文件
-    
-    Args:
-        race_audio_dir: race_audio目录路径
-        race_benchmark_path: race_benchmark.json文件路径
-        min_duration_seconds: 最小时长（秒），默认120秒=2分钟
-    """
-    
-    # 检查文件是否存在
-    if not os.path.exists(race_benchmark_path):
-        print(f"错误: 找不到benchmark文件: {race_benchmark_path}")
-        return
-    
-    if not os.path.exists(race_audio_dir):
-        print(f"错误: 找不到音频目录: {race_audio_dir}")
-        return
-    
-    # 加载benchmark JSON
-    print(f"加载benchmark文件: {race_benchmark_path}")
-    with open(race_benchmark_path, "r", encoding="utf-8") as f:
-        benchmark_data = json.load(f)
-    
-    print(f"原始benchmark包含 {len(benchmark_data)} 个QA对")
-    
-    # 获取所有唯一的音频文件路径并检查时长
-    audio_durations = {}
-    unique_audio_paths = set()
-    
-    for item in benchmark_data:
-        audio_rel_path = item["audio_path"]
-        unique_audio_paths.add(audio_rel_path)
-    
-    print(f"找到 {len(unique_audio_paths)} 个唯一音频文件")
-    print(f"开始检查音频文件时长（最小时长: {format_duration(min_duration_seconds)}）...")
-    
-    # 检查每个音频文件的时长
-    for i, audio_rel_path in enumerate(unique_audio_paths, 1):
-        audio_full_path = os.path.join(race_audio_dir, audio_rel_path)
-        
-        if not os.path.exists(audio_full_path):
-            print(f"警告: 文件不存在 {audio_full_path}")
-            audio_durations[audio_rel_path] = 0  # 不存在的文件时长设为0
-            continue
-        
-        # 获取音频时长
-        duration = get_audio_duration(audio_full_path)
-        audio_durations[audio_rel_path] = duration
-        
-        # 每处理100个文件显示一次进度
-        if i % 100 == 0:
-            print(f"已处理 {i}/{len(unique_audio_paths)} 个音频文件")
-    
-    # 过滤QA对，只保留音频时长满足要求的
-    filtered_benchmark_data = []
-    long_audio_count = 0
-    
-    for item in benchmark_data:
-        audio_rel_path = item["audio_path"]
-        duration = audio_durations.get(audio_rel_path, 0)
-        
-        if duration >= min_duration_seconds:
-            filtered_benchmark_data.append(item)
-    
-    # 统计结果
-    long_audio_paths = set()
-    for item in filtered_benchmark_data:
-        long_audio_paths.add(item["audio_path"])
-    
-    print(f"\n=== 过滤结果 ===")
-    print(f"原始QA对数量: {len(benchmark_data)}")
-    print(f"过滤后QA对数量: {len(filtered_benchmark_data)}")
-    print(f"保留的音频文件数: {len(long_audio_paths)}")
-    print(f"保留比例: {len(filtered_benchmark_data)/len(benchmark_data)*100:.1f}%")
-    
-    # 显示保留的音频文件统计
-    long_audio_info = []
-    for audio_path in long_audio_paths:
-        duration = audio_durations[audio_path]
-        long_audio_info.append({
-            "audio_path": audio_path,
-            "duration_seconds": duration,
-            "duration_formatted": format_duration(duration),
-            "qa_count": sum(1 for item in filtered_benchmark_data if item["audio_path"] == audio_path)
-        })
-    
-    # 按时长降序排序
-    long_audio_info.sort(key=lambda x: x["duration_seconds"], reverse=True)
-    
-    print(f"\n保留的音频文件详情（按时长降序）：")
-    for i, info in enumerate(long_audio_info[:10], 1):  # 只显示前10个
-        print(f"{i:2d}. {info['audio_path']}")
-        print(f"    时长: {info['duration_formatted']} ({info['duration_seconds']:.1f}秒)")
-        print(f"    QA对数量: {info['qa_count']}")
-    
-    if len(long_audio_info) > 10:
-        print(f"    ... 还有 {len(long_audio_info) - 10} 个音频文件")
-    
-    # ��存过滤后的benchmark到新文件
-    output_file = f"race_benchmark_filtered_{min_duration_seconds}s.json"
-    with open(output_file, "w", encoding="utf-8") as f:
-        json.dump(filtered_benchmark_data, f, ensure_ascii=False, indent=2)
-    
-    print(f"\n过滤后的benchmark已保存到: {output_file}")
-    
-    # 统计信息
-    all_durations = list(audio_durations.values())
-    if all_durations:
-        avg_duration = sum(all_durations) / len(all_durations)
-        max_duration = max(all_durations)
-        min_duration = min(all_durations)
-        
-        print(f"\n=== 统计信息 ===")
-        print(f"总音频文件数: {len(all_durations)}")
-        print(f"平均时长: {format_duration(avg_duration)} ({avg_duration:.1f}秒)")
-        print(f"最长时长: {format_duration(max_duration)} ({max_duration:.1f}秒)")
-        print(f"最短时长: {format_duration(min_duration)} ({min_duration:.1f}秒)")
-        print(f"长音频占比: {len(long_audio_paths)}/{len(all_durations)} ({len(long_audio_paths)/len(all_durations)*100:.1f}%)")
-    
-    return filtered_benchmark_data
-
-def main():
-    parser = argparse.ArgumentParser(description="根据音频时长过滤race_benchmark.json，生成新的benchmark文件")
-    parser.add_argument("--race_audio_dir", default="/root/autodl-tmp/project/Phi-4-multimodal-instruct/eval/test/race_audio", 
-                       help="race_audio目录路径 (默认: ./race_audio)")
-    parser.add_argument("--benchmark_json", default="./race_benchmark.json",
-                       help="race_benchmark.json文件路径 (默认: ./race_benchmark.json)")
-    parser.add_argument("--min_duration", type=str, default="2:00",
-                       help="最小时长，格式为 M:SS (默认: 2:00)")
-    
-    args = parser.parse_args()
-    
-    # 解析时长参数
-    try:
-        if ":" in args.min_duration:
-            parts = args.min_duration.split(":")
-            minutes = int(parts[0])
-            seconds = int(parts[1])
-            min_duration_seconds = minutes * 60 + seconds
-        else:
-            min_duration_seconds = int(args.min_duration)
-    except ValueError:
-        print(f"错误: 无效的时长格式 '{args.min_duration}'")
-        print("请使用格式 'M:SS' (如 2:00) 或直接输入秒数")
-        return
-    
-    print(f"过滤条件:")
-    print(f"  race_audio目录: {args.race_audio_dir}")
-    print(f"  benchmark文件: {args.benchmark_json}")
-    print(f"  最小时长: {args.min_duration} ({min_duration_seconds}秒)")
-    print()
-    
-    # 执行过滤
-    filter_race_benchmark_by_duration(
-        race_audio_dir=args.race_audio_dir,
-        race_benchmark_path=args.benchmark_json,
-        min_duration_seconds=min_duration_seconds
-    )
-
-if __name__ == "__main__":
-    main()

race_audio/.ipynb_checkpoints/race_long_audio_search_results_270s-checkpoint.json

@@ -1,14 +0,0 @@
-{
-  "search_criteria": {
-    "min_duration_seconds": 270,
-    "min_duration_formatted": "4:30"
-  },
-  "total_files_found": 0,
-  "long_audio_files": [],
-  "statistics": {
-    "total_audio_files": 1384,
-    "avg_duration_seconds": 94.15662933526019,
-    "max_duration_seconds": 253.125,
-    "min_duration_seconds": 27.425
-  }
-}

race_audio/.ipynb_checkpoints/serch_audio-checkpoint.py

@@ -1,191 +0,0 @@
-#!/usr/bin/env python3
-"""
-搜索race_audio中时长大于等于4分30秒的音频文件
-"""
-
-import json
-import os
-import librosa
-import soundfile as sf
-from collections import defaultdict
-import argparse
-
-def get_audio_duration(audio_path):
-    """获取音频文件的时长（秒）"""
-    try:
-        # 使用librosa获取音频时长（更快，只读取header）
-        duration = librosa.get_duration(path=audio_path)
-        return duration
-    except Exception as e:
-        try:
-            # 备选方案：使用soundfile
-            info = sf.info(audio_path)
-            duration = info.frames / info.samplerate
-            return duration
-        except Exception as e2:
-            print(f"无法读取音频文件 {audio_path}: {e2}")
-            return 0
-
-def format_duration(seconds):
-    """将秒数格式化为分:秒的形式"""
-    minutes = int(seconds // 60)
-    secs = int(seconds % 60)
-    return f"{minutes}:{secs:02d}"
-
-def search_long_audio_files(race_audio_dir, race_benchmark_path, min_duration_seconds=270):
-    """
-    搜索时长大于等于指定时间的音频文件
-    
-    Args:
-        race_audio_dir: race_audio目录路径
-        race_benchmark_path: race_benchmark.json文件路径
-        min_duration_seconds: 最小时长（秒），默认270秒=4分30秒
-    """
-    
-    # 检查文件是否存在
-    if not os.path.exists(race_benchmark_path):
-        print(f"错误: 找不到benchmark文件: {race_benchmark_path}")
-        return
-    
-    if not os.path.exists(race_audio_dir):
-        print(f"错误: 找不到音频目录: {race_audio_dir}")
-        return
-    
-    # 加载benchmark JSON
-    print(f"加载benchmark文件: {race_benchmark_path}")
-    with open(race_benchmark_path, "r", encoding="utf-8") as f:
-        benchmark_data = json.load(f)
-    
-    print(f"benchmark包含 {len(benchmark_data)} 个条目")
-    
-    # 获取所有唯一的音频文件路径
-    unique_audio_paths = set()
-    for item in benchmark_data:
-        audio_rel_path = item["audio_path"]
-        unique_audio_paths.add(audio_rel_path)
-    
-    print(f"找到 {len(unique_audio_paths)} 个唯一音频文件")
-    
-    # 检查每个音频文件的时长
-    long_audio_files = []
-    audio_durations = {}
-    
-    print(f"\n开始检查音频文件时长（最小时长: {format_duration(min_duration_seconds)}）...")
-    
-    for i, audio_rel_path in enumerate(unique_audio_paths, 1):
-        audio_full_path = os.path.join(race_audio_dir, audio_rel_path)
-        
-        if not os.path.exists(audio_full_path):
-            print(f"警告: 文件不存在 {audio_full_path}")
-            continue
-        
-        # 获取音频时长
-        duration = get_audio_duration(audio_full_path)
-        audio_durations[audio_rel_path] = duration
-        
-        # 检查是否满足时长要求
-        if duration >= min_duration_seconds:
-            long_audio_files.append({
-                "audio_path": audio_rel_path,
-                "full_path": audio_full_path,
-                "duration_seconds": duration,
-                "duration_formatted": format_duration(duration)
-            })
-        
-        # 每处理100个文件显示一次进度
-        if i % 100 == 0:
-            print(f"已处理 {i}/{len(unique_audio_paths)} 个音频文件")
-    
-    # 按时长降序排序
-    long_audio_files.sort(key=lambda x: x["duration_seconds"], reverse=True)
-    
-    # 显示结果
-    print(f"\n=== 搜索结果 ===")
-    print(f"找到 {len(long_audio_files)} 个时长大于等于 {format_duration(min_duration_seconds)} 的音频文件：")
-    
-    if long_audio_files:
-        print("\n详细列表（按时长降序排列）：")
-        for i, audio_info in enumerate(long_audio_files, 1):
-            print(f"{i:2d}. {audio_info['audio_path']}")
-            print(f"    时长: {audio_info['duration_formatted']} ({audio_info['duration_seconds']:.1f}秒)")
-            print(f"    完整路径: {audio_info['full_path']}")
-            print()
-    
-    # 统计信息
-    all_durations = list(audio_durations.values())
-    if all_durations:
-        avg_duration = sum(all_durations) / len(all_durations)
-        max_duration = max(all_durations)
-        min_duration = min(all_durations)
-        
-        print(f"=== 统计信息 ===")
-        print(f"总音频文件数: {len(all_durations)}")
-        print(f"平均时长: {format_duration(avg_duration)} ({avg_duration:.1f}秒)")
-        print(f"最长时长: {format_duration(max_duration)} ({max_duration:.1f}秒)")
-        print(f"最短时长: {format_duration(min_duration)} ({min_duration:.1f}秒)")
-        print(f"长音频占比: {len(long_audio_files)}/{len(all_durations)} ({len(long_audio_files)/len(all_durations)*100:.1f}%)")
-    
-    # 保存结果到JSON文件
-    result_data = {
-        "search_criteria": {
-            "min_duration_seconds": min_duration_seconds,
-            "min_duration_formatted": format_duration(min_duration_seconds)
-        },
-        "total_files_found": len(long_audio_files),
-        "long_audio_files": long_audio_files,
-        "statistics": {
-            "total_audio_files": len(all_durations),
-            "avg_duration_seconds": avg_duration if all_durations else 0,
-            "max_duration_seconds": max_duration if all_durations else 0,
-            "min_duration_seconds": min_duration if all_durations else 0
-        }
-    }
-    
-    output_file = f"race_long_audio_search_results_{min_duration_seconds}s.json"
-    with open(output_file, "w", encoding="utf-8") as f:
-        json.dump(result_data, f, ensure_ascii=False, indent=2)
-    
-    print(f"\n结果已保存到: {output_file}")
-    
-    return long_audio_files
-
-def main():
-    parser = argparse.ArgumentParser(description="搜索race_audio中时长大于等于指定时间的音频文件")
-    parser.add_argument("--race_audio_dir", default="/root/autodl-tmp/project/Phi-4-multimodal-instruct/eval/test/race_audio", 
-                       help="race_audio目录路径 (默认: ./race_audio)")
-    parser.add_argument("--benchmark_json", default="./race_benchmark.json",
-                       help="race_benchmark.json文件路径 (默认: ./race_benchmark.json)")
-    parser.add_argument("--min_duration", type=str, default="2:00",
-                       help="最小时长，格式为 M:SS (默认: 4:30)")
-    
-    args = parser.parse_args()
-    
-    # 解析时长参数
-    try:
-        if ":" in args.min_duration:
-            parts = args.min_duration.split(":")
-            minutes = int(parts[0])
-            seconds = int(parts[1])
-            min_duration_seconds = minutes * 60 + seconds
-        else:
-            min_duration_seconds = int(args.min_duration)
-    except ValueError:
-        print(f"错误: 无效的时长格式 '{args.min_duration}'")
-        print("请使用格式 'M:SS' (如 4:30) 或直接输入秒数")
-        return
-    
-    print(f"搜索条件:")
-    print(f"  race_audio目录: {args.race_audio_dir}")
-    print(f"  benchmark文件: {args.benchmark_json}")
-    print(f"  最小时长: {args.min_duration} ({min_duration_seconds}秒)")
-    print()
-    
-    # 执行搜索
-    search_long_audio_files(
-        race_audio_dir=args.race_audio_dir,
-        race_benchmark_path=args.benchmark_json,
-        min_duration_seconds=min_duration_seconds
-    )
-
-if __name__ == "__main__":
-    main()